Ninety‐Day Stroke Recurrence in Minor Stroke: Systematic Review and Meta‐Analysis of Trials and Observational Studies

This study is performed in agreement with the American Heart Association Journals' implementation of the Transparency and Openness Promotion Guidelines. This study adheres to the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses guidelines. The authors declare that all supporting data are available within the article. This study was registered with the International Prospective Register of Systematic Reviews (2020 CRD42020203309).

Articles were included if a cohort with minor stroke was described along with 90‐day stroke recurrence rates. Ischemic stroke was defined as a new neurologic deficit lasting at least 24 hours that was not attributable to a nonischemic cause, or a new neurologic deficit not attributable to a nonischemic cause and accompanied by neuroimaging evidence of new brain infarction. Acute minor stroke was defined by a score of 5 or less on the National Institutes of Health Stroke Scale (NIHSS; scores range from 0 to 42, with higher scores indicating greater deficits) either at the time of randomization (for randomized controlled trials [RCTs]) or final coding (for observational studies). Studies that defined minor stroke using modified Rankin Scale criteria or disposition from the emergency department were excluded.^{16, 19} Stroke recurrence was defined as sudden onset of a new focal neurological deficit with clinical or imaging evidence of infarction, or a rapid worsening of an existing focal neurologic deficit. We included RCTs and observational studies. We excluded review articles, studies with follow‐up duration other than 90 days, studies with narrow inclusion criteria, studies that did not report stroke recurrence or cohorts with minor stroke, study protocols, and opinion letters. If any cohorts overlapped, the larger cohort was kept, to prevent double counting of patients.²⁰ Studies were grouped by treatment type for the synthesis (dual antiplatelet therapy [DAPT] trial arms, non‐DAPT trial arms, and observational studies).

PubMed and reference lists of relevant papers were searched for studies published up to December 1, 2021, using terms including “mild stroke,” “minor stroke,” and “recurrence.” Full published papers and conference abstracts were considered. Reference lists of relevant articles and experts in the field were involved in identifying relevant studies.

The actual PubMed search strategy was (“mild stroke”[All Fields] OR “minor stroke”[All Fields]) AND (“recurrence”[All Fields] OR “recurrence”[MeSH Terms] OR “recurrence”[All Fields] OR “recurrences”[All Fields] OR “recurrencies”[All Fields] OR “recurrency”[All Fields] OR “recurrent”[All Fields] OR “recurrently”[All Fields] OR “recurrents”[All Fields]). No filters nor limits were used. There was no restriction on patient ages, study settings, or geographic areas.

Two independent reviewers (A.L., T.P.) decided whether each study met inclusion criteria and extracted data into a structured Excel spreadsheet. Titles and abstracts were initially scanned for relevance. Full text articles of the selected studies were then examined for inclusion criteria. If there were missing data in a relevant paper, the authors were contacted for unpublished data.

Data included study name, author, study type (RCT control, RCT experimental, observational), setting (single center or multicenter), cohort start and end years, publication year, NIHSS definition for minor stroke, cohort size, and number of events. The mean age of the entire study cohort was recorded. Median ages were converted to mean ages using $$$ =\frac{a+2m+b}{4} $$$, where m=median and a and b are the low and high end of the range.²¹ Aggregate percentage of patients with hypertension, diabetes, ischemic heart disease, and known atrial fibrillation were requested from the authors.

Assessment of risk of bias within studies was performed using the Joanna Briggs Institute checklist for studies reporting prevalence and cumulative incidence data²² by an independent reviewer (S.M.).

The primary outcome measure was the risk of 90‐day stroke recurrence (inclusive of both progression and recurrence). This was calculated as a proportion $$$ \left(\frac{\mathrm{number}\kern0.34em \mathrm{of}\ \mathrm{events}}{\mathrm{cohort}\kern0.34em \mathrm{size}}\right) $$$ and reported as a percentage. Ninety days was chosen as the follow‐up duration in line with major clinical trials.^{1, 2, 3, 4, 23}

Statistical analysis was performed in R with the help of the metafor²⁴ package. DAPT trial arms, non‐DAPT trial arms, and observational studies were analyzed both separately and pooled. Meta‐analysis was performed using the random effects model. Random effects was chosen over fixed effects to account for expected interstudy variability given the inclusion of observational studies in this meta‐analysis.²⁵ Raw proportions were used rather than inexact methods that can be challenging to interpret.²⁶ Three‐level meta‐analysis was chosen given the presumed dependency between each pair of effect sizes extracted per RCT.²⁰ The restricted maximum‐likelihood estimator method was used to estimate the amount of heterogeneity. DAPT trial arms, non‐DAPT trial arms, and observational studies were analyzed both separately and pooled. The inconsistency I² index,²⁷ the sum of the squared deviations from the overall effect and weighted by study size, was used to measure heterogeneity.²⁷ Metaregression was planned regardless of evidence for statistical heterogeneity; this decision was made to justify the inclusion of observational studies.²⁵ Three‐tiered study type (RCT experimental, RCT control, observational), 2‐tiered study type (RCT, observational), and time of onset to recruitment cutoff (12 hours, 24 hours, 48 hours, or none) were used as categorical moderators. Publication year was chosen as a covariate given reports of improving outcomes with contemporary management of minor stroke and TIA¹⁵ and TIA alone.^{11, 14, 28} Age, hypertension, diabetes, ischemic heart disease, and known atrial fibrillation were also tested as continuous moderators. Sensitivity analysis to compare various minor stroke definitions (NIHSS score ≤3, NIHSS score ≤4, and NIHSS score ≤5) and the exclusion of studies with n<500 to minimize the potential overestimating effect of studies with small sample sizes²⁹ was conducted to assess the robustness of the synthesized results.

The search yielded 14 eligible studies (6 RCTs and 8 observational studies) describing 45 462 patients. The Preferred Reporting Items for Systematic Reviews and Meta‐Analyses flow chart is shown in Figure S1. Regarding notable exclusions, only the largest Oxford Vascular Study (OXVASC) cohort was retained,³⁰ with previous overlapping cohorts excluded.^{15, 31} One major trial (THALES [Acute Stroke or Transient Ischemic Attack Treated with Ticagrelor and ASA (Aspirin) for Prevention of Stroke and Death]) was excluded as the follow‐up ended at 30 days.³² One major registry (tiaregistry.org)¹⁶ and 1 observational study (NORTHSTAR [North West of England Transient Ischaemic Attack and Minor Stroke]¹⁹) chose a modified Rankin Scale score of ≤1 as the definition of minor stroke. The FASTER (Fast Assessment of Stroke and Transient Ischemic Attack to Prevent Early Recurrence) trial, although using NIHSS score ≤3, was excluded as the minor stroke subset data could not be obtained from the authors.³³ One observational study defined minor stroke at discharge, that is, “any ischemic stroke (International Classification of Diseases, Tenth Revision, Canadian Version [ICD‐10‐CA] codes H341, I63* [except I636], I64*, I676) that was discharged home with or without support directly from the emergency department”³⁴ and was excluded due to not meeting the NIHSS score ≤5 on admission criteria and unavailability of minor stroke subset data.

The data contain 45 462 patients. Characteristics are summarized in the Table.

Three observational studies were single center, and all other studies were multicenter. Most studies chose to define minor stroke as NIHSS score ≤3 (9/14=64%). Less common definitions were NIHSS score ≤5 (2/14=14%), NIHSS score ≤4 (1/14=7%), or NIHSS score ≤2 (1/14=7%). The mean NIHSS score was 3.3±0.8. The TARDIS (Triple Versus Guideline Antiplatelet Therapy to Prevent Recurrence After Acute Ischemic Stroke or Transient Ischemic Attack) trial did not have an NIHSS upper limit for recruitment,⁵ but the authors were able to provide data for NIHSS score ≤3. For the purposes of the subgroup analysis, the TARDIS experimental arm (aspirin, clopidogrel, and dipyridamole) was classified as a DAPT arm and the control arm (clopidogrel alone or aspirin and dipyridamole) a non‐DAPT arm. For 9 studies, additional unpublished data were obtained from the authors to clarify outcomes for minor stroke in isolation. This included data obtained from the National Institute of Neurological Disorders and Stroke and from AstraZeneca. The minor stroke data from SOCRATES (Acute Stroke or Transient Ischemic Attack Treated With Aspirin or Ticagrelor and Patient Outcomes)² were provided with the requirement that Chinese patients were removed.

The pooled 90‐day stroke recurrence for minor stroke was 8.6% (95% CI, 6.5–10.7)—see Figure. Recurrence was lowest in DAPT arms (6.3%, 95% CI, 4.5, 8.0) when compared with non‐DAPT arms (7.2%, 95% CI, 4.7–9.6) and observational studies 10.6% (95% CI, 7.0–14.2) (Figure). Test of moderators using study type as a 3‐tiered categorical moderator (RCT experimental, RCT control, observational) demonstrated significant difference between study types (Q_M=21.59, df=2, P<0.001). Study type as a 2‐tiered categorical moderator (RCT, observational) trended to significant (Q_M=3.24, df=1, P=0.07). Time of onset to recruitment as a categorical moderator (12 hours, 24 hours, 48 hours, or none) was not significant (Q_M=2.42, df=3, P=0.49). Metaregression with continuous variables demonstrated a significant reduction in the estimate with each subsequent year of publication 0.60% (95% CI, 0.09–1.1, P=0.02). No significant trend was observed with age (β=−0.003, 95% CI, −0.009 to 0.02, P=0.23), hypertension (β=−0.001, 95% CI, −0.003 to 0.001, P=0.48), diabetes (β=0.001, 95% CI, −0.002 to 0.005, P=0.51), ischemic heart disease (β=−0.001, 95% CI, −0.007 to 0.005, P=0.63), known atrial fibrillation (β=0.001, 95% CI, −0.002 to 0.004, P=0.35), or percentage antiplatelet treatment (redundant predictor). Defining minor stroke with a lower NIHSS threshold made no difference—NIHSS score ≤3: 8.6% (95% CI, 6.0–11.1), NIHSS score ≤4: 8.4% (95% CI, 6.1–10.6), as did excluding studies with n<500%–7.3% (95% CI, 5.5–9.0). Assessment of individual study bias demonstrated low risk overall, with none of the selected studies scoring a “no” on any of the 9 domains assessed. This is provided in the Table S1.

Our study was made possible through the generous release of previously unpublished data from randomized trials. However, a caveat is that the data on minor stroke from SOCRATES² trial did not contain information on Chinese patients, and the FASTER data could not be obtained. As such our study does not contain all possible data. Selection bias between trials and observational studies exist due to trials having stringent inclusion and exclusion criteria compared with observational studies. Furthermore, not all minor strokes are the same, as those with large vessel occlusion having thrombectomy or thrombolysis are generally excluded from these studies.⁴¹ In addition, the use of penumbral imaging to select candidates for reperfusion (and therefore exclude from the cohorts described in this paper) was not reported, adding a further unmeasured confounder. Another potential source of bias is measurement bias with regard to diagnosis of stroke recurrence, which can differ between RCTs and observational studies. Finally, it would be ideal to perform this analysis with individual patient data.⁴² However, this would prove difficult, as some of the authors contacted stated that the data were no longer accessible due to expiry of ethics approval or archiving.